CODEWITHSUNDEEP
rmachine-learningtime-seriestraining-dataarima
Tuey
Tuey

Reputation: 1

Training set and ARIMA plot isn't working?

I'm trying to plot my training data using ARIMA with a dataset that has the dates from 2021/02/28 - 2022/03/22. But when I plot my training ARIMA, it plots 2021-2022, leaves a gap in the graph and starts predicting from 2027-2028 instead .

What am i doing wrong?

My code:

#load dataset [390 rows x 15 columns]
df = read_csv("data.csv")
#convert to time series and select the 4th column 
ts <- ts(df[1:390,4], start = c(2021, 02), end = c(2022, 03), frequency = 52)

#train data (80-20)
train <- ts(ts[1:311], start = c(2021, 02), frequency = 52)
test <- ts(ts[312:390], start = c(2021, 011),frequency = 52)

#get summary
myarima <- auto.arima(train)
summary(myarima)

#plot training ARIMA data and original dataset
fc_arima1 <- forecast(myarima, 52)
autoplot(fc_arima1)

my attempt: weird arima plot

Upvotes: 0

Views: 164

Answers (1)

Rob Hyndman
Rob Hyndman

Reputation: 31820

Your ts object has 390 observations, but the last 336 are missing because you have set end to be part way through the data set. With frequency of 52, there are only 54 observations between the start of c(2021,02) and the end of c(2022,04).

You should also use either the window or subset functions to form the training and test sets, so you don't have to re-specify the time series attributes.

The following should work. I've assumed your start date was wrong but your end date was correct:

#convert to time series and select the 4th column 
ts <- ts(df[1:390,4], end = c(2022, 04), frequency = 52)

#train data (80-20)
train <- subset(ts, end=311)
test <- subset(ts, start=312)

#fit model
myarima <- auto.arima(train)
summary(myarima)

#plot training ARIMA data and original dataset
fc_arima1 <- forecast(myarima, 52)
autoplot(fc_arima1)

Upvotes: 1

Related Questions